From d136db5171a3fee8a53195470b48ca9ef5e6d109 Mon Sep 17 00:00:00 2001 From: seungrokj Date: Fri, 24 Apr 2026 12:15:05 +0900 Subject: [PATCH 1/3] Update GLM5 FP8 atom config: new image, add TP=4, set gpu-memory-utilization Co-Authored-By: Claude Sonnet 4.6 --- .github/configs/amd-master.yaml | 4 +++- benchmarks/single_node/glm5_fp8_mi355x_atom.sh | 2 ++ perf-changelog.yaml | 6 ++++++ 3 files changed, 11 insertions(+), 1 deletion(-) diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index 0cd82774e..9cdd8797d 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -362,7 +362,7 @@ glm5-fp8-mi355x-sglang-mtp: - { tp: 8, conc-start: 4, conc-end: 64, spec-decoding: mtp } glm5-fp8-mi355x-atom: - image: rocm/atom:rocm7.2.1-ubuntu24.04-pytorch2.9.1-atom0.1.2.post + image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post model: zai-org/GLM-5-FP8 model-prefix: glm5 runner: mi355x @@ -373,10 +373,12 @@ glm5-fp8-mi355x-atom: - isl: 1024 osl: 1024 search-space: + - { tp: 4, conc-start: 4, conc-end: 256 } - { tp: 8, conc-start: 4, conc-end: 256 } - isl: 8192 osl: 1024 search-space: + - { tp: 4, conc-start: 4, conc-end: 256 } - { tp: 8, conc-start: 4, conc-end: 256 } glm5.1-fp4-mi355x-sglang: diff --git a/benchmarks/single_node/glm5_fp8_mi355x_atom.sh b/benchmarks/single_node/glm5_fp8_mi355x_atom.sh index 31bc8b25f..036346af3 100644 --- a/benchmarks/single_node/glm5_fp8_mi355x_atom.sh +++ b/benchmarks/single_node/glm5_fp8_mi355x_atom.sh @@ -39,6 +39,7 @@ fi # Start GPU monitoring (power, temperature, clocks every second) start_gpu_monitor +MEM_FRAC_STATIC=0.9 set -x pip install -U transformers @@ -47,6 +48,7 @@ python3 -m atom.entrypoints.openai_server \ --server-port $PORT \ -tp $TP \ --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \ + --gpu-memory-utilization $MEM_FRAC_STATIC \ --default-chat-template-kwargs '{"enable_thinking": false}' \ --trust-remote-code \ > $SERVER_LOG 2>&1 & diff --git a/perf-changelog.yaml b/perf-changelog.yaml index ddc6409c2..da81d6df9 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1733,3 +1733,9 @@ - "TP=2 and TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths" - "Add --max-num-seqs and --gpu-memory-utilization 0.9 to server launch" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1043 + +- config-keys: + - glm5-fp4-mi355x-atom + description: + - "" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/ From 8a1620cf1a379294efa5007b4cb6ada3de5da71f Mon Sep 17 00:00:00 2001 From: seungrokj Date: Fri, 24 Apr 2026 12:19:21 +0900 Subject: [PATCH 2/3] Fix perf-changelog entry for GLM5 FP8 atom config Co-Authored-By: Claude Sonnet 4.6 --- perf-changelog.yaml | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index da81d6df9..4195d8564 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1734,8 +1734,11 @@ - "Add --max-num-seqs and --gpu-memory-utilization 0.9 to server launch" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1043 -- config-keys: - - glm5-fp4-mi355x-atom +- config-keys: + - glm5-fp8-mi355x-atom description: - - "" + - "Update GLM-5 FP8 MI355X ATOM benchmark: new image, add TP=4, set gpu-memory-utilization" + - "Image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post" + - "Add TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths" + - "Add --gpu-memory-utilization 0.9 to server launch" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/ From 66c42b42da74cca5e452f47e34ae2b514b90e6bc Mon Sep 17 00:00:00 2001 From: seungrokj Date: Fri, 24 Apr 2026 12:20:16 +0900 Subject: [PATCH 3/3] Add PR link to perf-changelog for GLM5 FP8 atom config Co-Authored-By: Claude Sonnet 4.6 --- perf-changelog.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 4195d8564..35a98bcf9 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1741,4 +1741,4 @@ - "Image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post" - "Add TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths" - "Add --gpu-memory-utilization 0.9 to server launch" - pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/ + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1126